How to get to UF Shands CDC for hospotal scrubs

Go to the east end of the hospital where the gift shop is.
Take the elevator for the basement.
The door with the buzzer is the place you are looking for.

better estimator...but why?

Instead of Monte Carlo integrating the Hellinger divergence directly, if I use the Phi-divergence form (modified to be symmetric), then both the convergence speed is faster and the power of the hypothesis test is faster...at least in the Poisson vs PTST case.

Old:


New:

Total variation (Phi-divergence)

Synchrony detectors

My paper on a new point process model got rejected because of the estimation method I introduced was far from complete. I was wondering if I should try to decompose signals into precisely timed components and others. That would be more practical, and I can use the existing synchrony detection methods on multiple trials to detect those events. Hmmm

Incognito mode of Chrome

I love the incognito mode key combination of google chrome browser, "Ctrl-Shift-N".
It instantly forks a fresh browser process that does not share any cookies with my other browser windows. I use it whenever I see a suspicious link that someone sent or is 'tinyurl'ed. These can be dangerous because they can sometimes use the already logged in brower/tab's cookies to do stuff. For example, the viral twitter applications use this to send messages to people you follow, without you noticing (you just click a wrong button). I can't wait to see how Chrome will be advancing with plugins.

Q: Integrative management of personal information

My data are not integrated nor very well organized. I use a somewhat distributed and categorized storage system, but I really wish I had something that can do the following. Does anybody know such a software system?

Requirements
1. backed up: nothing is more frustrating than loosing all your information at once
2. secure: I want to be able to dump all the sensitive information in my brain to somewhere
3. control I want to be the person who controls my information, not bound by some EULA that I won't read.
4. widely used data format: in case I want to migrate the data to some other format in the future
5. keyboard only navigation: I don't like moving my hand to the mouse too often
6. simple hierarchical organization + tags
7. fast text search
8. versioning

Memming's AMD Rig @ Home

I just built a new desktop computer for myself a couple of days ago. This is my first AMD machine. (I owned more than 10 intel machines and others)
Here's the sepc:

• AMD Phenom II x2 550 BE 3.1 GHz (unlocked to x4)


• ASUS M4A785TD-M EVO motherboard (BIOS 0604, 8/4/9)


• G.SKILL 4GB (2 x 2GB) 240-Pin DDR3 SDRAM DDR3 1600 (PC3 12800) Dual Channel Kit F3-12800CL9D-4GBNQ


• SeaSonic S12 Energy Plus SS-550HT 550W ATX12V V2.3 / EPS12V V2.91 SLI Certified CrossFire Ready 80 PLUS


• Rosewill R5601-BK 0.8mm Japanese Cold Rolled Steel Screw-less Dual 120mm Fans ATX Mid Tower


• SAMSUNG Spinpoint F3 HD502HJ 500GB 7200 RPM SATA 3.0Gb/s 3.5" Internal Hard Drive


• Sony Optiarc 24X DVD/CD Rewritable Drive Black SATA Model AD-7240S-0B


• Dell E207WFP widescreen LCD 20" 1680 x 1050

I unlocked the disabled cores using the unleashing mode from recent BIOS from ASUS. Also I increased the memory clock to 800MHz (CAS is 11). I must say the system performance is AWESOME!

Currently I'm using Windows 7 Ultimate RC from MSDN. I'm pretty impressed with the OS. So far I had two problems: Vim self-extracting installer fails (I used the zipped version which fails to install the context menu), and some java swing application seems to hang.

I know regular expressions

About 12 years ago, I learned regular expressions (also learned regular language in undergrad automata class). I'm just so proud of myself everytime I use complicated replacements. I think everybody should learn it. At least, I won't say everybody should learn vim.

5ms vs 2ms with Silverman's rule

Averaged squared Hellinger divergence computed for the same dataset, just different kernel sizes.

Uncoupled oscillators

Can uncoupled oscillators with common input and output process information? For example, audio filter banks are constructed in similar manner. Each frequency components are separately processed and then combined at the end. The key is the phase distribution in case of oscillators. That's all they have. But individual oscillators can only see their own one. What type of interaction between the input and individual phase would make this a useful computational device or memory device?

Inter trial stationarity

When conducting multiple trials of an experiment, we hope to be able to assume that the trials are independent of each other and the experimental conditions are stationary. But how can we show this? Some sort of statistical test would be nice...but that would still require independence, wouldn't it? So many things to assume to release a little bit of assumption. What an irony.

SugarSync!

SugarSync is a very nice and fast file system sync tool.

It also has a web-based MUSIC PLAYER, so that I don't need to install the client and sync all the files first. When I am working on somebody else's computer this is superb. Also it syncs with my blackberry. Nice!

Point process space and spike train space

In ITL (information theoretic learning; @CNEL), instead of using a kernel between two data samples, an expectation of product of probability estimate is used as a kernel. It's called the cross information potential (CIP). Therefore, the kernel is really between statistics of two datasets, not just two sample points. Of course, when a single data point is used to estiamte the PDF, it becomes just like the normal kernel method. The inner product defined by CIP uses the statistics of the data (which can be equivalently represented by the mean of samples in the RKHS). [See Xu, Paiva, Park, Principe 2009]
Same thing is growing in my research on spike train lately. Instead of defining a kernel between two spike trains, I would first estimate point processes from a set of spike train and then create an inner product based on CIP or other cpd kernels.
The question is where this would be useful. The single trial approximation is not so useful because the structure over the trials is lost. It would be same as the mCI RKHS we proposed. [See Paiva, Park, Principe 2009] The biggest problem is how to get a nonparametric estimator for point processes. They are very high dimension!!

Donation to TeX

I donated $20 each to TUG and MikTeX. I love TeX/LaTeX!
@ $q(t \cdot \pi)$

Massive renaming task

On windows I recommend the tool, "ReNamer" by Denis Kozlov.
You can find it on http://www.den4b.com.

Removing non-stationarity

Today I removed 100 trials out of 600 at the beginning which are the most non-stationary ones. And the distribution of Fano factors drastically changed.

Previously, out of 600 experiments (each with 600 trials), 268 had Fano factor greater than 1, but after removing the first 100 trials, I got 155 experiments.

Memory dump from COSYNE day 2

This is my second time coming to COSYNE, so I know the 7:30am-11:30pm schedule will exhaust my short-term memory very quickly, so I am trying to write down my interpretation and ideas about the interesting talks while in the dinner break. Hopefully, I'll be able to understand it later on. :)

Earl K. Miller from MIT used decoding techniques to determine the order of information processing in the brain areas LIP, FEF , and PRC. The firing rate increases almost immediately after the stimulation, but it does not encode the decision that is made until some time after. It seems to be an interesting technique, but it heavily relies on the ability to detect such coding. Extraction of relevant information using conditional entropy might be a good thing to try (Sohan!). He also proposes that LFP oscillation cycles in the beta range might cause the internal attention shift cycle. Furthermore, he shows that the averaging window time lock to the onset is poorer than using the last two LFP cycles. However, an audience asked an excellent question that the LFP does not look like sine waves unless band-pass filtered and smoothed. Also, the action potential and LFP are always in a egg-or-chicken dilemma.

Vikaas Sohal from Stanford showed that a cortical microcircuit with a piramidal neuron and a fast spiking inhibitory neuron enhances gamma-band oscillation using various fancy biological methods. He had three different stimulation pattern that he injected via dynamic clamp or light-induced current to mimic EPSCs: non-rhythmic, and two different frequencies. When I asked him if the non-rhythmic stimulation was frozen Poisson process, but he said it was not frozen, and not Poisson. I shall ask him (if I can find him free somewhere during the conference) how he can be sure that the reduction of variability is not coming from reduction of variability in the input signal.

Antonio Rangel from Caltech talked about neuroeconimocs. He described a diffusion based model to describe how internal value and attention can fit various psychophysical results. Take home message: the longer you look at things you like, the more likely you would choose it. (The longer you look at disgusting things, the less likely that you would eat it)

Robert Wilson from UPenn talked about the change point problem; to detect where the signal was non-stationary in a simple case. The algorithm has to distinguish variation due to noise and jumps; to do so it has to estimate the noise, and the frequency of change. Using a Bayesian approach, they were able to come up with almost exact inference algorithm. I wonder if autocorrentropy estimator can do similar things without the complicated algorithm (I think like an engineer too...sometimes)

Misha Tsodyks talked about things that most of them I already read, but listening in person is a world of a difference. He was trying to promote the computational role of short-term plasticity (his dynamics synpase model) via the notion of population spike. One work I wasn't aware of was the work on auditory coding which intrigued me because I did some reading on Meddis hair cell model recently.
After working on the BORNs for a while, the work on working memory in non-spiking state which looked like a trivial idea (although published on nature, I think) sounds very interesting as well. The past is encoded in the distribution of phases instead of facillitation variables of the synapses in our case. :D
Overall, I didn't like his talk too much, because although related to many of my research problems, he does not have a strong experimental support of what he describes in model exist in the real brain.

I also had a random idea about how to better model the inter-population spike-interval using some sort of ghost of saddle type of dynamics, instead of completely using the depression time constant which is around the range of 1 second while we need something n the order of 4~10 seconds.

Cori Bargmann from Rockefeller University gave a nice talk about the neuromodulatory mechanisms that are not described in the full 'connectome' of C. elegance. Using diverse genetic manipulations and calcium imaging, her group showed some amazing results of how seemingly feedforward network had feedback modulation through modulatory peptides.

P.S. Due to a stupid mistake I missed a flight and as a consequence the keynote speech and posters of day 1 which was about olfactory coding. :'(

Comparing likelihoods of heterogeneous models

Given a dataset X, which model is the best model to fit it?
In a parametric family of models, often this problem is tackled by maximum likelihood (estimation of the parameters). However if the family of models has too much freedom, it may overfit, the same way function approximation and regression does. If a model A predicts the dataset will be X and only X, given the dataset X, it would obviously have the maximum likelihood. It is like doing a pdf estimation with kernel density estimation with a Dirac delta kernel.
How can we avoid this situation?
Since it is caused by the model's lack of generalization ability, one obvious way is to use a test set (or cross validation). If the likelihood value for the test set and training set does not differ significantly, it is not overfitting.
Can we compare the likelihoods of two MLEs on two different model families that does not overfit the data? I think so.

Noise and regularization in function approximation

Problem: function approximation using standard kernel method, which kernel should I use? Given a finite training set of data, you can easily overfit by choosing a narrow kernel, which is equivalent to just memorizing the data points.
If we know there's noise in the system, we can do a little better in terms of choosing the right kernel by means of regularization. The idea is that similar points should be mapped to similar points through the function; it's a generalization of continuous functions in a topological space setting. The noise in the input space can be used to define the similarity in that space. In case of real valued functions, the system is of the form: Y = f(x+N), and you want to approximate with y = sum alpha_i K(x_i, x). The noise variable inside the true function makes your conditional expectation smoother than the original.

However, additive noise does not work the same way. It does not smooth the function at all! Only when the function is linear they would be equivalent.

Same argument works when the output space is class labels or clusters.

Rozin's talk inspired me

A psychologist Paul Rozin came to UF to give a talk on the smell and taste seminar last week. He started off by talking about the importance of first describing phenomena in contrary to explaining them. I have forgotten about looking for novel phenomena that were not properly described in the literature! I have been working too hard to model and explain things, and not paid attention to the spirit of description.
The reason he brought this up is because he is kind of non-mainstream psychologist. Many of his research ends up not being able to be published. According to him, this is mostly because they are not subject of main stream academia.

The main topic was in human, there are many things that should be naturally avoided, but you learn to like it later in life; such as hotness (almost no animal likes hot food, except for maybe parrots), bitterness, sadness, pain, scariness, disgusting blue cheese, etc. A new born babies would not like any of these, and at some age, perhaps due to social context, they develop a preference for them!

The seminar was hilarious, and it was definitely the most entertaining and inspiring talks in 2 years for me.

P.S. He also talked about his experience in one of the best restaurants in the world, El Bulli in Spain.